Skip to content

Add oneDNN, TensorArray support, Bugfix

Pre-release
Pre-release
Compare
Choose a tag to compare
@jxt1234 jxt1234 released this 20 Apr 09:53
· 612 commits to master since this release
1.1.2
d766c6d

Features:

  1. Add OneDNN for Convolution Opt (In Test)
  2. Add Codegen for Fuse Op (In Test)
  3. Add TensorArray support (In Test)

Speed up / Bugfix:

  1. Support more binary op for ARMv8.2
  2. Optimize Int8 compute for AVX2 and Neon
  3. Fix bug for region fuse for spectial case
  4. Fix bug for Vulkan Convolution Winograd run error for feature map is large
Assets 2
Loading